Presentation is loading. Please wait.

Presentation is loading. Please wait.

© Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 1 Genetic Algorithms (in 1 Slide) l GA: based on an analogy to biological evolution l Each.

Similar presentations


Presentation on theme: "© Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 1 Genetic Algorithms (in 1 Slide) l GA: based on an analogy to biological evolution l Each."— Presentation transcript:

1 © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 1 Genetic Algorithms (in 1 Slide) l GA: based on an analogy to biological evolution l Each rule is represented by a string of bits l An initial population is created consisting of randomly generated rules l Based on the notion of survival of the fittest, a new population is formed to consists of the fittest rules and their offspring l The fitness of a rule is represented by its classification accuracy on a set of training examples l Offspring are generated by crossover and mutation l GA’s are a general search/optimization method, not just a classification method. This can be contrasted with other methods

2 © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 2 Ensemble Methods l Construct a set of classifiers from the training data l Predict class label of previously unseen records by aggregating predictions made by multiple classifiers l In Olympic Ice-Skating you have multiple judges? Why?

3 © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 3 General Idea

4 © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 4 Why does it work? l Suppose there are 25 base classifiers –Each classifier has error rate,  = 0.35 –Assume classifiers are independent –Probability that the ensemble classifier makes a wrong prediction: –Practice has shown that even when independence does not hold results are good

5 © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 5 Methods for generating Multiple Classifiers l Manipulate the training data –Sample the data differently each time –Examples: Bagging and Boosting l Manipulate the input features –Sample the featurres differently each time  Makes especially good sense if there is redundancy –Example: Random Forest l Manipulate the learning algorithm –Vary some parameter of the learning algorithm  E.g., amount of pruning, ANN network topology, etc.  Use different learning algorithms

6 © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 6 Background l Classifier performance can be impacted by: –Bias: assumptions made to help with generalization  "Simpler is better" is a bias –Variance: a learning method will give different results based on small changes (e.g., in training data).  When I run experiments and use random sampling with repeated runs, I get different results each time. –Noise: measurements may have errors or the class may be inherently probabilistic

7 © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 7 How Ensembles Help l Ensemble methods can assist with the bias and variance –Averaging the results over multiple runs will reduce the variance  I observe this when I use 10 runs with random sampling and see that my learning curves are much smoother –Ensemble methods especially helpful for unstable classifier algorithms  Decision trees are unstable since small changes in the training data can greatly impact the structure of the learned decision tree –If you combine different classifier methods into an ensemble, then you are using methods with different biases  You are more likely to use a classifier with a bias that is a good match for the problem  You may even be able to identify the best methods and weight them more

8 © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 8 Examples of Ensemble Methods l How to generate an ensemble of classifiers? –Bagging –Boosting l These methods have been shown to be quite effective l A technique ignored by the textbook is to combine classifiers built separately –By simple voting –By voting and factoring in the reliability of each classifier

9 © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 9 Bagging l Sampling with replacement l Build classifier on each bootstrap sample l Each sample has probability (1 – 1/n) n of being selected (about 63% for large n) –Some values will be picked more than once l Combine the resulting classifiers, such as by majority voting l Greatly reduces the variance when compared to a single base classifier

10 © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 10 Boosting l An iterative procedure to adaptively change distribution of training data by focusing more on previously misclassified records –Initially, all N records are assigned equal weights –Unlike bagging, weights may change at the end of boosting round

11 © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 11 Boosting l Records that are wrongly classified will have their weights increased l Records that are classified correctly will have their weights decreased Example 4 is hard to classify Its weight is increased, therefore it is more likely to be chosen again in subsequent rounds

12 © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 12 Class Imbalance l Class imbalance occurs when the classes in the distribution are very unevenly distributed –Examples would include fraud prediction and identification of rare diseases l If there is class imbalance, accuracy may be high even if the rare class is never predicted –This could be okay, but only if both classes are equally important  This is usually not the case. Typically the rare class is more important  The cost of a false negative is usually much higher than the cost of a false positive –The rare class is designated the positive class

13 © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 13 Confusion Matrix l The following abbreviations are standard: –FP: false positive –TP: True Positive –FN: False Negative –TN: True Negative Predicted Class Actual Class + - + - TPFN FPTN

14 © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 14 Classifier Evaluation Metrics l Accuracy = (TP + TN)/(TP + TN + FN + FP) l Precision and Recall –Recall = TP/(TP + FN)  Recall measure the fraction of positive class examples that are correctly identified –Precision = TP/(TP + FP)  Precision is essentially the accuracy of the examples that are classified as positive –We would like to maximize precision and recall  They are usually competing goals  These measure are appropriate for class imbalance since recall explicitly addresses the rare class

15 © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 15 Classifier Evaluation Metrics Cont. l One issue with precision and recall is that they are two numbers, not one –That makes simple comparisons more difficulty and it is not clear how to determine the best classifier –Solution: combine the two l F-measure combines precision and recall –The F 1 measure is defined as:  (2 x recall x precision)/(recall + precision)

16 © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 16 Cost-Sensitive Learning l Cost-sensitive learning will factor in cost information –For example, you may be given the relative cost of a FN vs. FP –You may be given the costs or utilities associated with each quadrant in the confusion matrix l Cost-sensitive learning can be implemented by sampling the data to reflect the costs –If the cost of a FN is twice that of a FP, then you can increase the ratio of positive examples by a factor of 2 when constructing the training set

17 © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 17 More on autonomous vehicles l DARPA Grand Challenges. $1,000,000 prize. –Motivation: by 2015 1/3 of ground military forces autonomous. –2004: 150 mile race through the Mojave desert. No one finished. CMU’s car made it the farthest at 7.3 miles –2005: Same race. 22 of 23 surpassed the best distance from 2004. Five vehicles completed the course. Stanford first, CMU second. Sebastian Thrun leader for Stanford team. –2005 Grand Challenge Video2005 Grand Challenge Video

18 © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 18 More DARPA Grand Challenge l 2007 Urban Challenge –60 mile urban area course to be completed in 6 hours.  Must obey all traffic laws and avoid other robot cars –Urban Challenge VideoUrban Challenge Video

19 © Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 19 Google Autonomous Vehicle l Google “commercial” videovideo l Second Google driverless car videovideo l Alternative future autonomous “vehicles” –videovideo


Download ppt "© Tan,Steinbach, Kumar Introduction to Data Mining 4/18/2004 1 Genetic Algorithms (in 1 Slide) l GA: based on an analogy to biological evolution l Each."

Similar presentations


Ads by Google